A session is more than chat history
A durable coding-agent session can retain the task, selected model profile, mode, permission policy, plan, useful workspace notes, and a bounded conversation transcript. That makes it possible to resume a real task without pretending every previous token needs to be sent again.
Persistence should be intentional. A session needs enough state to continue responsibly, but it should not silently preserve credentials or grow into an unsearchable archive of stale context.
Checkpoints make interruptions recoverable
Coding work is interrupted by editor restarts, failed provider requests, network changes, and simple context switches. A checkpoint records where the agent was, what it had observed, and what still needs verification so the next run can recover without guessing.
- Task and implementation-plan status
- Selected provider, model, and agent mode
- Relevant workspace notes and changed-file state
- Bounded conversation history and recent tool outcomes
Resume with verification, not blind continuation
A resumed agent should re-check the workspace before it acts. Files may have changed, a branch may have moved, or a previous command may not have completed. The right behavior is to use the stored session as a starting point, then inspect current evidence.
Truss treats plans, workspace memory, and sessions as durable runtime state while still keeping tool activity and Git review visible when work resumes.
Next steps
Put it to work in your own workspace.
Pick the surface that fits your day, then connect a model you control.
FAQ
Questions developers ask before switching.
Should an AI coding agent remember everything forever?
No. Keep durable task and workspace state, but bound conversation history and re-check live files before resuming work.
What is an agent checkpoint?
A checkpoint is saved state that helps an agent resume after an interruption, including task progress, relevant context, and recent outcomes.